# Demographics and Japanification: Who's Next?

## Abstract

We test whether demographic structure predicts "Japanification" --- the syndrome of simultaneously low growth, low inflation, and low interest rates --- using Fair-Dominguez polynomial age-distribution variables in a panel of 194 countries from 1990 to 2024. The answer is: weakly yes, but unstably so. On a cleaned two-component index (growth and inflation, with shock-year exclusions and five-year rolling averages), the first demographic polynomial is significant (Z₁ = 1.179, SE = 0.395, p = 0.003), but the persistent channel is disinflation, not growth: on five-year rolling averages, all three Z terms significantly predict lower inflation (Z₁ = −3.25, p < 0.001) while no Z term predicts growth (Z₁ p = 0.56). In raw annual data, demographics sometimes correlate with growth, but that relationship is transitory and does not survive rolling-average smoothing. Old-age dependency ratio spline regressions identify a growth-drag threshold at 15%, above which the association strengthens (p = 0.002). The demographic-Japanification link exhibits a dramatic structural break around the Global Financial Crisis: Z₁ reverses sign from +2.89 (p < 0.001) pre-GFC to −1.46 (p < 0.001) post-GFC, a pattern that survives balanced-panel, OECD/non-OECD, income-quartile, and global-factor controls --- ruling out entry/exit of countries as the primary explanation. However, the QE/non-QE split (Section 6.4) reveals that policy-regime heterogeneity already present pre-GFC contributes to the break, suggesting the structural break reflects a combination of changing global conditions and shifting composition of policy regimes. On a restricted sample using only 10-year government bond yields (eliminating rate measurement heterogeneity), demographics do predict interest rates (Z₁ = 53.1, p = 0.011), but this rate channel disappears post-GFC. The effect is concentrated in middle-income countries (~$8,000 GDP per capita) and is absorbed by working-age share controls. Conditional projections diverge dramatically across estimation periods: under pre-GFC coefficients, 10 of 24 focus economies cross the Japanification threshold by 2050; under post-GFC coefficients, none do.

---

## 1. Introduction

Japan's experience since the early 1990s --- persistent low growth, near-zero inflation, and interest rates at or below the zero lower bound --- has become a reference point for advanced economies confronting population aging. The term "Japanification" has entered both the academic and policy lexicon, typically referring to the simultaneous occurrence of these three conditions as a chronic state rather than a cyclical downturn. As Europe's demographics increasingly resemble Japan's of two decades ago, and as Korea, China, and Thailand undergo rapid aging, the question of whether Japanification is a general consequence of demographic structure has gained urgency.

The theoretical mechanisms are well-established. The lifecycle hypothesis (Modigliani and Brumberg, 1954) predicts that aging populations save more in aggregate as workers prepare for longer retirements, depressing interest rates. Carvalho, Ferrero, and Nechio (2016) formalize how aging shifts the natural rate of interest downward. Summers (2014) frames this as "secular stagnation" --- the possibility that aging economies face chronically insufficient demand. Goodhart and Pradhan (2020) controversially argue that the relationship may reverse as the ratio of consumers to producers rises, making aging eventually inflationary.

Despite these theoretical contributions, no formal cross-country econometric test has examined whether demographic structure --- specifically, the Fair-Dominguez polynomial representation of the age distribution --- systematically predicts a composite Japanification syndrome. Individual channels have been studied: demographics and growth (Fair and Dominguez, 1991), demographics and inflation (Juselius and Takáts, 2015), demographics and interest rates (Carvalho et al., 2016). But the joint prediction --- that the same age-structure variables should simultaneously predict low growth, low inflation, and low rates --- has not been tested.

This paper fills that gap, and the results are more nuanced than the simple aging-causes-Japanification narrative suggests. We make five contributions.

First, we construct a continuous Japanification index from standardized growth, inflation, and interest rate components, along with binary indicators and persistence-weighted variants, enabling formal panel estimation.

Second, we estimate the relationship between Fair-Dominguez demographic polynomials and a cleaned rolling index using pooled GLS with AR(1) correction across 164 countries from 1990 to 2024. The baseline result is significant (Z₁ p = 0.003, Z₂ p = 0.011), with an R² of 0.059.

Third, we document a sharp structural break around the Global Financial Crisis. On the clean rolling index, Z₁ reverses from +2.89 (p < 0.001) pre-GFC to -1.46 (p < 0.001) post-GFC. This is not a composition artifact: it survives a balanced panel of 156 countries present in both periods, holds in both OECD and non-OECD subsamples, persists across income quartiles, and is robust to controlling for GDP-weighted world growth as a global factor.

Fourth, we show that the income-level heterogeneity in this relationship is the opposite of what canonical theory predicts. The demographic effect is significant only for middle-income countries (second income quartile, ~$8,000 GDP per capita), not for the rich aging economies where Japanification is most discussed. This effect is substantially attenuated by controlling for the working-age population share (53% reduction), indicating that it partly captures the demographic dividend ending --- the macroeconomic slowdown when working-age share peaks and declines --- rather than aging at the top of the age distribution alone.

Fifth, we provide conditional projections through 2060 based on UN population estimates under three scenarios (pre-GFC, post-GFC, and full-sample coefficients). The scenarios diverge dramatically: under pre-GFC coefficients, 10 of 24 focus economies cross the Japanification threshold by 2050; under post-GFC or full-sample coefficients, none do. This divergence underscores the fragility of demographic Japanification projections.

The remainder of this paper proceeds as follows. Section 2 reviews the related literature. Section 3 describes the data and index construction. Section 4 presents the methodology. Section 5 reports the main results. Section 6 investigates the structural break. Section 7 examines income-level heterogeneity. Section 8 provides projections. Section 9 reports robustness tests. Section 10 discusses implications and concludes.


## 2. Literature Review

### 2.1 Japanification: Concept and Mechanisms

The term "Japanification" lacks a single formal definition but generally refers to a macroeconomic syndrome characterized by three simultaneous conditions: low real GDP growth (typically below 2%), low inflation (below 2%, sometimes deflation), and low nominal interest rates (near the zero lower bound). Japan has exhibited all three since the early 1990s, following the bursting of its asset price bubble and amid rapid population aging.

Several theoretical channels connect demographics to this syndrome:

*Growth channel.* A declining working-age population directly reduces labor input and, through dependency effects, may reduce capital accumulation. Aksoy et al. (2019) estimate that a one percentage point increase in the old-age dependency ratio reduces GDP growth by 0.1--0.2 percentage points.

*Inflation channel.* The lifecycle framework predicts that a high ratio of dependents (both young and old) to workers is inflationary, as consumption exceeds production. Juselius and Takáts (2015) document a U-shaped relationship: both youth and old-age dependency are associated with higher inflation, while a large working-age share is disinflationary. However, Goodhart and Pradhan (2020) argue that aging is inflationary through labor scarcity, predicting a reversal of the post-1990 low-inflation environment as the baby boom generation retires.

*Interest rate channel.* Carvalho, Ferrero, and Nechio (2016) show that aging depresses the natural rate of interest as retirees dissave but are offset by workers increasing precautionary savings in response to longer life expectancy. Rachel and Summers (2019) attribute approximately 1.7 percentage points of the secular decline in real interest rates to demographic factors.

### 2.2 Cross-Country Evidence

The empirical literature on demographics and macroeconomic outcomes is extensive for individual channels but sparse for the composite syndrome. Higgins (1998) established that age-distribution variables predict current account balances in cross-country panels. Koomen and Wicht (2020) embedded demographic polynomials in the IMF's External Balance Assessment framework. Our companion papers (Papers 1--3) extend this analysis to bilateral flows, financial openness interactions, and causal identification.

For the Japanification composite, the closest precedent is the descriptive literature identifying countries "at risk" based on aging indicators (e.g., the Bank for International Settlements' Annual Reports). However, this literature relies on ad hoc thresholds rather than formal econometric testing. Bouis et al. (2013) examine Japan's experience econometrically but focus on a single-country time series rather than cross-country panel variation.

### 2.3 The Fair-Dominguez Polynomial Approach

We employ the Fair and Dominguez (1991) polynomial constraint, which parameterizes age-group coefficients as a cubic function of age-group index, reducing 17 age-group parameters to 3 summary variables (Z₁, Z₂, Z₃). This approach has become standard in the demographic-macro literature, balancing parsimony with flexibility in capturing lifecycle patterns. The zero-sum restriction ensures that a uniform age distribution has no net effect. For our analysis, this approach allows direct comparison with the current account regressions in our companion papers.


## 3. Data and Index Construction

### 3.1 Panel Assembly

We draw from the panel constructed for our multilateral current account project, which assembles data from ten international sources including the IMF World Economic Outlook, UN World Population Prospects (2024 revision), World Bank World Development Indicators, and the Chinn-Ito KAOPEN index.

From this panel, we select observations for 1990--2024, the period during which Japanification became relevant as a macroeconomic concept. We require non-missing values for real GDP growth, inflation, and the three demographic polynomial variables (Z₁, Z₂, Z₃). Inflation is taken from the IMF WEO as the primary source, with World Bank CPI inflation as fallback (103 observations filled). Interest rates follow a hierarchy: 10-year government bond yields where available (23 countries), then policy rates, then lending rates. This hierarchy mixes fundamentally different instruments: government bond yields reflect term premia and sovereign credit risk; policy rates reflect central bank stance; lending rates embed bank-specific margins. We address this heterogeneity in two ways: our primary analysis uses a two-component index (growth and inflation only), and Appendix A reports results on a restricted sample using only 10-year government bond yields as a homogeneous rate measure. Inflation is winsorized at the 1st and 99th percentiles to limit the influence of hyperinflation episodes.

The resulting panel contains 6,436 observations across 194 countries, of which 4,271 have an interest rate measure.

### 3.2 Japanification Index

We construct several index variants, with the cleaned rolling two-component index as our primary measure:

**Binary indicator.** A country-year is classified as "Japanified" if real GDP growth is below 2% and inflation is below 2%. Under this definition, 10.0% of country-years qualify (644 of 6,436). A stricter variant additionally requiring rates below 2% classifies 1.0% (65 observations).

**Continuous two-component index.** We standardize growth and inflation to zero mean and unit variance across the panel, then average with sign inversion so that higher values indicate more Japanification:

$$J_{it}^{2c} = -\frac{1}{2}\left(\frac{g_{it} - \bar{g}}{\sigma_g} + \frac{\pi_{it} - \bar{\pi}}{\sigma_\pi}\right)$$

We use the two-component index (growth and inflation only) as the basis for our primary measure, avoiding the rate measurement heterogeneity that would arise from splicing 10-year bond yields, policy rates, and lending rates into a single variable (see Section 3.1). A three-component version including rates is reported in Appendix A on a restricted sample.

**Shock filter and winsorization.** Country-years with real GDP growth below -10% are excluded from the index (119 observations across 71 countries, including war zones, COVID outliers, and financial collapses). The remaining index values are winsorized at the 1st and 99th percentiles ([-2.60, 0.92]).

**Primary index: cleaned rolling average.** Our primary dependent variable is a five-year moving average (minimum 3 periods) of the shock-filtered, winsorized two-component index. This captures the persistent nature of Japanification as a chronic state rather than a single-year event, and eliminates contamination from transitory shocks. The rolling clean index has 6,001 observations across 194 countries (mean = 0.000, SD = 0.369).

### 3.3 Descriptive Statistics

Japan ranks sixth in our country rankings by mean Japanification index (0.29), behind several small states (FSM, Palau, Puerto Rico) but ahead of Italy (0.26), Switzerland (0.25), Germany (0.24), Finland (0.24), and France (0.24). The top-20 most Japanified country-years in the raw index include Macau 2020 (COVID contraction, growth -54.3%), Kuwait 1990--91 (Gulf War, growth -41.0%), and Ukraine 2022 (Russian invasion, growth -28.8%), illustrating that extreme index values often reflect event shocks rather than structural Japanification. Our shock filter (Section 3.2) removes 119 such observations; the rolling average further smooths transitory spikes.


## 4. Methodology

### 4.1 Estimation

We estimate pooled GLS with AR(1) error correction, following the EBA methodology employed in our companion papers. The model is:

$$J_{it} = \gamma_1 Z_{1,it} + \gamma_2 Z_{2,it} + \gamma_3 Z_{3,it} + \beta' X_{it} + u_{it}$$

where $u_{it} = \rho u_{i,t-1} + \varepsilon_{it}$, and the panel-wide $\rho$ is estimated iteratively via Cochrane-Orcutt. Controls $X$ include fiscal balance (% GDP), KAOPEN, log relative output per worker, and lagged NFA/GDP. This specification deliberately mirrors the current account regressions from Paper 1, enabling direct comparison of demographic coefficients across dependent variables.

### 4.2 Component Regressions

To decompose which channel drives the composite result, we regress each Japanification component separately on the same demographic and control variables:

$$g_{it} = \gamma^g Z_{it} + \beta^g X_{it} + u^g_{it}$$
$$\pi_{it} = \gamma^\pi Z_{it} + \beta^\pi X_{it} + u^\pi_{it}$$
$$r_{it} = \gamma^r Z_{it} + \beta^r X_{it} + u^r_{it}$$

### 4.3 Threshold and Nonlinearity Tests

We test for OADR thresholds using spline regressions with candidate knots at 15%, 20%, 25%, and 30%, and for life expectancy convexity using a quadratic specification. Rolling 15-year window estimation tests for time-varying relationships.


## 5. Main Results

### 5.1 Baseline Estimates

Table 1 reports the baseline results for the cleaned rolling two-component Japanification index (4,049 observations, 164 countries after requiring non-missing controls). Results are from the shock-filtered, winsorized, five-year rolling average specification described in Section 3.2.

All three demographic polynomials are significant: Z₁ = 1.179 (SE = 0.395, p = 0.003), Z₂ = −0.150 (SE = 0.059, p = 0.011), and Z₃ = 0.006 (SE = 0.002, p = 0.019). The R² is 0.059 --- modest but comparable to the early specifications in the current account literature before financial openness interactions were added. Log relative output per worker is the strongest predictor (−0.051, p < 0.001), followed by fiscal balance (−0.002, p < 0.001). KAOPEN is not significant (0.007, p = 0.118).

These results represent an improvement over the raw (non-rolling, non-cleaned) two-component index, where Z₁ was larger in magnitude (1.204) but with similar significance (p = 0.019). The shock filter and rolling average sharpen the demographic signal by removing transitory events that contaminate the relationship.

KAOPEN interaction terms (Z × KAOPEN) are not significant in any specification --- a notable contrast with the current account regressions from Paper 1, where all three interactions were highly significant (p < 0.005). Financial openness does not appear to mediate the demographic-Japanification link.

### 5.2 Component Regressions

Decomposing the index into its rolling components reveals a more nuanced pattern than the raw-data regressions suggest:

- *Z → Growth (rolling):* No demographic term is significant (Z₁ = 0.52, p = 0.56; Z₂ p = 0.36; Z₃ p = 0.34). On five-year rolling averages, the transitory demographic-growth relationship washes out.
- *Z → Inflation (rolling):* All three demographic terms are highly significant (Z₁ = −3.25, p < 0.001; Z₂ = 0.46, p < 0.001; Z₃ = −0.018, p < 0.001). Aging demographics persistently predict lower inflation, consistent with the standard disinflationary effect of population aging and inconsistent with the Goodhart-Pradhan reversal thesis.

This pattern reverses the raw-data finding (where growth was marginally significant and inflation was null) and carries an important interpretation: the transitory demographic signal operates through growth (cyclical), while the persistent signal operates through inflation (structural). The five-year rolling average correctly captures the persistent component relevant to the Japanification concept.

The interest rate channel is addressed in Appendix A. On the heterogeneous rate measure (mixing bond yields, policy rates, and lending rates), demographics are actually significant but with the wrong sign for the Japanification narrative (Z₁ = −43.9, p = 0.010 --- aging predicts *lower* rates, not higher Japanification). Restricting to the 23 countries with 10-year government bond yields produces a similar result: Z₁ = 53.1 (p = 0.011), Z₂ = −5.96 (p = 0.031), Z₃ = 0.19 (p = 0.071). Demographics do predict interest rates when measured homogeneously --- but only pre-GFC (see Section 6).

**OECD vs. non-OECD channel decomposition.** The aggregate channel results mask important income-group heterogeneity. The inflation channel is significant *only* in OECD countries (Z₁ = −1.86, p = 0.002); in non-OECD countries, the inflation channel is not significant (p = 0.10). Conversely, the rate channel is significant *only* in non-OECD countries (Z₁ = −0.71, p = 0.017); in OECD countries, the rate channel is not significant (p = 0.33). The growth channel is null in both subsamples (OECD p = 0.23, non-OECD p = 0.27). This decomposition has a natural interpretation: in OECD economies, where inflation expectations are better anchored and monetary policy frameworks are more credible, the demographic channel operates through persistent disinflation. In non-OECD economies, where financial markets are less developed and rates reflect broader macroeconomic conditions, demographics affect the rate channel instead. Neither group shows a significant growth channel on rolling averages, reinforcing the conclusion from the full sample that the growth association is transitory.

### 5.3 OADR Growth-Drag Threshold

Old-age dependency ratio spline regressions on the clean rolling index identify a threshold at 15%, where both below and above coefficients are significant (below = 1.31, p < 0.001; above = 1.38, p < 0.001). The 15% knot produces the best fit (R² = 0.097), above the baseline Z-polynomial R² of 0.059. At higher knots (25% and 30%), the above-threshold coefficient loses significance, consistent with a threshold that activates at relatively low levels of aging.

We label this a "demographic transition threshold" rather than a "Japanification threshold" because the OADR spline on the growth component directly shows a similar pattern (below-15% coefficient = −5.69, p < 0.001; above-15% = −1.27, p = 0.081), while the inflation component does not display a comparable threshold structure. As documented in Section 5.2, the persistent demographic channel on rolling averages operates through disinflation rather than growth. The spline's explanatory power comes primarily from the growth component: the growth-specific spline shows a similar threshold structure (below-15% = −5.69***, above-15% = −1.27*, p = 0.081), while the inflation component does not display a comparable discontinuity. The 15% threshold likely marks the point at which the demographic dividend ends and the composite index begins to rise, driven primarily by transitory growth effects rather than persistent disinflationary pressure, and should not be interpreted as the point at which the full Japanification syndrome activates.

### 5.4 Life Expectancy

Life expectancy displays significant convexity (LE² p < 0.001) on the clean rolling index, with a composite turning point at approximately 59.3 years for the full sample. Below this threshold, increasing life expectancy is associated with *less* Japanification (through development effects); above it, further increases are associated with *more* Japanification. The earlier estimate of 41.8 years reflected the rate channel specifically; the composite turning point is driven by the growth channel, where the turning point is 60.6 years.

Critically, the turning point rises monotonically with income. Splitting by income quartile: Q1 (poorest, ~$2,500 GDP/capita) = 64.8 years; Q2 (~$8,200) = 66.2 years; Q3 (~$18,300) = 72.6 years; Q4 (richest, ~$47,700) = 87.5 years, though the Q4 quadratic is not statistically significant --- the relationship is essentially linear for the richest countries. For the OECD subsample, the turning point is 85.0 years, with both LE and LE² significant (p < 0.03 and p < 0.04 respectively). This means that for OECD economies, Japanification risk keeps rising with life expectancy almost throughout the entire observed range --- there is no point at which further longevity gains relieve the Japanification association.

The income gradient has an intuitive interpretation. In poorer countries, the turning point arrives earlier because life expectancy gains initially reflect development (reducing Japanification through growth) but the development dividend fades at lower LE levels. In richer countries, the development effect persists much longer --- extending life expectancy from 75 to 85 in an advanced economy continues to reflect health system improvements and productivity gains rather than pure aging drag. Only at very high life expectancy levels (above 85 in OECD) would the aging effect begin to dominate, and for the richest quartile the quadratic is not even significant, suggesting the relationship remains approximately linear throughout the observed range.

### 5.5 Age-Group Profile

Direct estimation using 17 age-group bins (rather than the polynomial constraint) reveals which age groups drive the Japanification association. The most significant positive coefficients (more Japanification) are for the 30--34 group (7.2, p = 0.021) and the 55--59 group (6.6, p = 0.023). The 80+ group is positive but only marginally significant (5.2, p = 0.077). The most significant negative coefficients (less Japanification) are for the 65--69 group (−9.4, p = 0.006), the 10--14 group (−7.7, p = 0.012), and the 50--54 group (−7.3, p = 0.016).

This profile does not match the simple lifecycle prediction. The strong negative coefficient on the 65--69 group --- the early-retirement cohort --- is counterintuitive under a pure aging story. It may reflect that countries with disproportionately large 65--69 cohorts are at an earlier stage of aging (the baby boom generation just reaching retirement), where the economic effects of the demographic transition have not yet fully materialized, while countries with large 80+ cohorts represent more advanced aging. We note that the 17-bin specification involves joint estimation of many correlated regressors and should be treated as exploratory. With 17 bins, approximately one significant coefficient at p < 0.05 would be expected by chance alone; we do not interpret individual bin coefficients, only broad patterns. The key takeaway is that the early-retirement cohort (65--69) is strongly associated with *less* Japanification, while working-age (30--34, 55--59) and advanced-aging (80+) groups show positive associations.

### 5.6 Cross-Validation

As a sanity check, we re-estimate the standard Z → CA/GDP regression on the Japanification sample (1990--2024). However, in the expanded 140-country panel, no Z term is significant (Z₁ = 7.26, p = 0.56; Z₂ p = 0.63; Z₃ p = 0.70). This likely reflects the broader developing-country composition of this sample compared to the more restrictive sample used in Paper 1, where fiscal balance (p < 0.001) and log relative output per worker (p < 0.001) dominate the current account specification. The weaker current account results on this sample do not undermine the Japanification findings, as the two dependent variables capture different phenomena.


## 6. The Post-GFC Structural Break

### 6.1 Rolling Window Evidence

Rolling 15-year window estimation reveals a dramatic pattern. For windows ending before 2012, Z₁ is positive (coefficients 1.9--4.2), significant at the 1% level through 2011 and losing significance as the window begins to include post-GFC data. For windows ending after 2016, Z₁ turns negative. The transition occurs in windows spanning the Global Financial Crisis.

The formal pre/post split on the clean rolling index confirms this: pre-GFC (1990--2007), all three Z terms are significant (Z₁ = 2.89, p < 0.001; Z₂ = −0.45, p < 0.001; Z₃ = 0.019, p < 0.001). Post-GFC (2009--2024), all three Z terms are significant but with *reversed signs* (Z₁ = −1.46, p < 0.001; Z₂ = 0.21, p < 0.001; Z₃ = −0.008, p = 0.001). This is not merely a loss of significance but a sign reversal: post-GFC, aging demographics are associated with *less* Japanification, not more.

**Balanced panel test.** To rule out entry/exit of countries as an explanation, we restrict to 156 countries present in both periods (with at least 5 observations each). The break is identical: pre-GFC Z₁ = 2.89 (p < 0.001), post-GFC Z₁ = −1.46 (p < 0.001). This rules out differential country inclusion as the source, though the QE split (Section 6.4) shows that policy-regime heterogeneity already present pre-GFC contributes to the break.

**OECD vs. non-OECD.** The break manifests differently across income groups. OECD demographics were never significant pre-GFC (Z₁ = −0.47, p = 0.67) but become strongly significant post-GFC with reversed sign (Z₁ = −5.08, p < 0.001). Non-OECD countries show the canonical pattern: significant pre-GFC (Z₁ = 3.14, p < 0.001), reversed post-GFC (Z₁ = −1.68, p < 0.001). The pre-GFC full-sample result was driven entirely by non-OECD countries.

**Income quartile splits.** Pre-GFC significance is concentrated in Q2 (lower-middle income, Z₁ = 6.25, p < 0.001) and Q3 (upper-middle income, Z₁ = 3.54, p = 0.014). Post-GFC, Q3 and Q4 show the strongest reversals (Q3 Z₁ = −2.84, p = 0.003; Q4 Z₁ = −2.16, p = 0.003). Q1 is not significant in either period, and Q4 pre-GFC is also null (Z₁ = 0.83, p = 0.25).

**Global factor control.** Adding GDP-weighted world growth as a control does not alter the pattern. Z₁ survives pre-GFC (3.12, p < 0.001) with world growth also significant (−0.010, p < 0.001). Post-GFC, Z₁ remains reversed (−1.45, p < 0.001) while world growth is null (p = 0.21). The domestic demographic signal is not absorbed by global conditions.

### 6.2 Which Component Broke?

Component-by-component rolling windows (Section 4b) identify the growth channel as the locus of the break. The Z₁ coefficient for growth averages approximately −25 in early windows (windows ending 2004--2009, all significant at 5%) versus +5 to +8 in late windows (none significant). The inflation channel was significant in the earliest windows (Z₁ around −130, p < 0.001 for windows ending 2004--2007) but lost significance entirely after 2011, consistent with globalization and inflation-targeting regimes suppressing the cross-country inflation signal. The rate channel showed a similar pre-GFC significance (Z₁ around −80, p < 0.02) that disappeared by 2011.

Formal pre/post-GFC splits on each channel confirm that the collapse is universal. Growth: pre-GFC Z₁ = −3.05 (p < 0.05), post-GFC Z₁ = +0.69 (NS). Inflation: pre-GFC Z₁ = −3.21 (p < 0.001), post-GFC Z₁ = −0.35 (NS). Rate: pre-GFC Z₁ = −0.96 (p < 0.05), post-GFC Z₁ = −0.19 (NS). Every individual channel loses significance post-GFC --- the structural break is not confined to a single component but reflects a broad decoupling of demographics from all three Japanification channels.

This suggests that demographics predicted growth slowdowns before the GFC but ceased to do so afterward. The inflation channel's early significance and subsequent collapse is notable --- it parallels the well-documented "death of Phillips curve" phenomenon in the post-GFC era. One interpretation is that post-crisis policy interventions --- fiscal stimulus, unconventional monetary policy, structural reforms --- disrupted the demographic-growth link. Another is that the pre-GFC relationship captured the convergence dynamics of transition and emerging economies, which are less relevant in the post-GFC era.

### 6.3 Ruling Out Mechanical Explanations

We test whether the break is an artifact of the interest rate channel losing variation at the zero lower bound (Story B). The two-component rolling windows (growth and inflation only, no rates) directly confirm this: Z₁ averages +3.5 in early windows ending 2004--2009 (100% significant at p < 0.001) and turns negative in windows ending after 2014, with Z₁ around −0.9 to −1.1 in the latest windows (significant at 5% in windows ending 2020--2022). **The break persists without the rate channel**, rejecting the mechanical explanation.

### 6.4 QE vs. Non-QE Countries

Splitting by whether countries conducted quantitative easing reveals a subtlety: the two groups had *opposite* Z₁ signs even before the GFC. QE countries (primarily advanced economies that later adopted unconventional monetary policy) showed Z₁ = −2.57 (p = 0.037) pre-GFC, while non-QE countries showed Z₁ = +3.65 (p < 0.001). Post-GFC, QE countries flip to Z₁ = +2.86 (p = 0.13, not significant) while non-QE countries reverse to Z₁ = −1.07 (p = 0.045).

This reveals that the "break" is partly a composition effect. The pre-GFC full-sample result was driven by the large non-QE group, where aging demographics did predict Japanification. The advanced QE economies always showed a different (inverted) pattern, which dominates in later periods as more country-years enter the post-crisis era.

### 6.5 ZLB and Fiscal Controls

Zero lower bound proximity does not explain the break. ZLB interaction terms (Z × near-ZLB dummy) are not significant in either the post-GFC sample or the full sample. Fiscal impulse (change in fiscal balance) is itself a significant predictor of Japanification (p < 0.001 in both periods) but does not restore the Z signal post-GFC. The break appears to be genuine rather than an artifact of omitted policy variables.

### 6.6 Interpreting the Break

The structural break admits multiple interpretations, and the decomposition results constrain but do not eliminate the candidates:

*Global contagion.* Post-GFC, low rates and low inflation became universal across advanced economies regardless of demographic profile. Unconventional monetary policy (QE) compressed interest rates globally through spillovers and carry trades. The "global savings glut" (Bernanke, 2005) may have evolved from a demographically driven phenomenon into a self-sustaining equilibrium maintained by central bank balance sheets and institutional demand for safe assets. However, controlling for GDP-weighted world growth does not restore the pre-GFC relationship (Section 6.1), which argues against a simple global-factor explanation.

*Convergence-era artifact.* The pre-GFC period included the convergence dynamics of transition and emerging economies, which may have amplified the demographic-growth correlation. The concentration of pre-GFC significance in non-OECD countries (Section 6.1) is consistent with this interpretation.

*Policy regime change.* Post-crisis fiscal stimulus, unconventional monetary policy, and structural reforms may have disrupted the demographic-growth link. Fiscal balance remains a strong predictor in both periods, suggesting that active policy matters. However, controlling for fiscal impulse does not restore the Z signal post-GFC (Section 6.5).

*Composition shift.* Country entry/exit is ruled out by the balanced-panel test: the same 156 countries show the break. However, the QE split (Section 6.4) reveals that the two policy-regime groups (QE and non-QE countries) had *opposite* Z₁ signs even pre-GFC, and the changing relative weight of QE-type observations contributes to the aggregate break. The break is therefore linked to policy-regime heterogeneity that was already present pre-GFC, not solely to a discrete structural change.

We cannot definitively adjudicate among these candidates. The most likely explanation is a combination: the pre-GFC relationship captured the demographic dividend dynamics of converging economies (primarily non-OECD), while the post-GFC reversal reflects a regime in which policy and institutional factors dominate the macroeconomic landscape, overwhelming individual demographic signals. The sign reversal --- not merely a loss of significance --- suggests active countervailing forces, not merely noise.


## 7. Income-Level Heterogeneity

### 7.1 The Puzzle

The income split from Section 5 produced a counterintuitive result: demographic polynomials are significant for low-income countries (all p < 0.001) but not for high-income countries (all p > 0.80). Japanification is discussed primarily in the context of rich aging economies, yet the empirical relationship is concentrated in poorer ones.

### 7.2 Ruling Out Restricted Variation

One explanation is statistical: perhaps rich countries are all similarly aged, providing insufficient variation to identify the effect. The variation diagnostic rejects this. High-income countries have *more* Z variation than low-income countries (SD ratios of 1.1--1.8 across the three polynomial terms). The between-country variation in Z₁ is also comparable (1.08 vs. 0.97). Restricted variation does not explain the result.

### 7.3 The Demographic Dividend Explanation

The decisive test is adding working-age population share to the low-income regression. When we do so, the Z₁ coefficient drops from 3.66 (p < 0.001) to 1.71 (p = 0.29) --- a 53% reduction. Working-age share itself enters positively but is not significant at the 5% level (4.50, p = 0.12). The attenuation is substantial though less complete than in the prior 69-country sample, reflecting the broader developing-country composition where the demographic polynomial captures additional variation beyond working-age share.

This means the "Japanification" signal in low-income countries is almost entirely about the working-age share peaking and declining --- the end of the demographic dividend --- rather than about aging at the top of the age distribution. Countries whose working-age share is shrinking experience lower growth and lower inflation, which mechanically increases the Japanification index. But this is a development phenomenon, not an aging phenomenon.

Intriguingly, for high-income countries, adding working-age share makes Z₁ *negative* and significant (−2.02, p = 0.024). Once the working-age share effect is controlled for, aging in rich countries is associated with *less* Japanification, not more. However, this should not be read as confirmation of Goodhart and Pradhan's (2020) thesis that aging is inflationary through labor scarcity. The rolling component regressions (Section 5.2) show that demographics predict inflation in the standard disinflationary direction (Z₁ = −3.25, p < 0.001 on rolling inflation), not in the inflationary direction that Goodhart and Pradhan predict. The high-income negative Z₁ conditional on working-age share operates through the growth channel: once the working-age share effect is netted out, aging in rich countries is associated with *less* growth drag, perhaps because these economies have adapted through automation, immigration, or institutional flexibility. The mechanism is a growth-side story, not the inflationary reversal that Goodhart and Pradhan hypothesize.

### 7.4 Matched Sample

Restricting both income groups to the overlapping OADR range to ensure demographic comparability, the disparity persists: low-income Z₁ = 4.43 (p < 0.001) versus high-income Z₁ = −1.07 (p = 0.21). The pooled matched regression with income interactions confirms a highly significant differential (Z₁ × high-income = −7.20, p < 0.001). This is not an artifact of different aging levels.

### 7.5 Income Quartile Analysis

Finer-grained analysis by income quartile localizes the effect to the second quartile (median GDP per capita ~$8,200 PPP): Z₁ = 6.66 (p < 0.001). The poorest quartile (~$2,500) is not significant (p = 0.71), nor are the third (~$18,300, p = 0.80) or fourth (~$47,700, p = 0.44). The demographic-Japanification link is a middle-income phenomenon, concentrated in countries at the inflection point of the demographic transition.

### 7.6 Interaction with the Structural Break

When the income split is crossed with the pre/post-GFC split, low-income countries are strongly significant pre-GFC (Z₁ = 5.98, p < 0.001) while high-income countries are not (Z₁ = 1.20, p = 0.18). Neither is significant post-GFC (high-income p = 0.38, low-income p = 0.28). The income disparity is a pre-GFC phenomenon, not a general feature of the data.


## 8. Projections

### 8.1 Methodology and Caveats

We compute projected Japanification indices through 2060 under three sets of estimated coefficients --- pre-GFC (1990--2007), post-GFC (2009--2024), and full sample --- using UN World Population Prospects demographic projections and holding control variables at their last observed values. Given the documented structural break, presenting a single set of projections would be misleading. The three-scenario approach provides maximum transparency about how sensitive projections are to the choice of estimation period.

The Japan threshold for the clean rolling index is 0.365, chosen because it corresponds to Japan's rolling clean index value circa 2000 --- the period when Japan's Japanification syndrome was first widely recognized as a persistent regime rather than a cyclical downturn.

### 8.2 Country Timelines

Table 7 reports projected 2050 index values and threshold crossing years under all three scenarios:

| Country | Pre-GFC 2050 | Full 2050 | Post-GFC 2050 | Crossing (Pre) | Crossing (Post) |
|---------|-------------|-----------|---------------|----------------|-----------------|
| Japan | 1.080 | 0.351 | 0.112 | Already | Never |
| Korea | 0.990 | 0.351 | 0.200 | ~2040 | Never |
| Italy | 0.998 | 0.328 | 0.140 | ~2040 | Never |
| Spain | 0.851 | 0.301 | 0.170 | ~2040 | Never |
| Greece | 0.769 | 0.292 | 0.165 | ~2040 | Never |
| Germany | 0.688 | 0.237 | 0.135 | ~2040 | Never |
| France | 0.563 | 0.205 | 0.127 | ~2040 | Never |
| China | 0.424 | 0.270 | 0.222 | ~2050 | Never |
| Thailand | 0.469 | 0.248 | 0.165 | ~2050 | Never |
| Canada | 0.363 | 0.177 | 0.166 | ~2040 | Never |
| Poland | 0.355 | 0.197 | 0.262 | ~2060 | Never |
| Singapore | 0.181 | 0.145 | 0.271 | ~2060 | Never |
| USA | 0.215 | 0.132 | 0.178 | Never | Never |
| UK | 0.309 | 0.161 | 0.162 | Never | Never |
| India | -0.306 | 0.078 | 0.165 | Never | Never |
| Brazil | -0.019 | 0.130 | 0.207 | Never | Never |
| Nigeria | -0.386 | -0.051 | -0.022 | Never | Never |

*Threshold: Japan's 2000 rolling clean index level (0.365). Pre-GFC coefficients: 10/24 cross by 2050. Post-GFC and full-sample coefficients: 0/24 cross.*

The divergence across scenarios is extreme. Under pre-GFC coefficients, Japan's projected 2050 index exceeds 1.0 and Korea approaches it (0.99). Under post-GFC coefficients, no country crosses the threshold and the highest projected value is Singapore (0.27). The full-sample estimates, which average across the pre- and post-GFC regimes, produce intermediate values that are still insufficient for any country to cross (though Japan and Korea both reach 0.35, just below the 0.365 threshold). This divergence reflects the fundamental fragility of demographic Japanification projections: they depend entirely on which estimation period one trusts, and the post-GFC evidence suggests that the demographic relationship has reversed.

### 8.3 Global Trajectory

Under the pre-GFC scenario, the share of focus countries exceeding the onset threshold rises from near-zero in 2000 to approximately 42% (10 of 24) by 2050. Under the full-sample and post-GFC scenarios, no country crosses. The general equilibrium concern --- that many major economies simultaneously entering Japanification could accelerate global interest rate compression --- is therefore conditional on the pre-GFC relationship continuing to hold, which the post-GFC evidence suggests it does not.

### 8.4 KAOPEN Scenarios

Financial openness scenarios (fully open KAOPEN = 2.4 versus closed KAOPEN = −1.9) produce large differences for financially less open economies. China's projected index shifts from 0.06 under current openness to 0.37 if fully open, reflecting the positive coefficient on KAOPEN in the Japanification regression. For already-open economies (Japan, Germany, USA), the differences are small. These KAOPEN scenario effects are illustrative and should be interpreted cautiously given KAOPEN's limited precision in the baseline regression (p = 0.118).


## 9. Robustness

### 9.1 Country Exclusions

Dropping Japan has negligible effect on the baseline (Z₁ = 1.25, p = 0.017 versus baseline 1.18, p = 0.003). Dropping CCA countries strengthens the result (Z₁ = 1.64, p = 0.002; all Z terms p < 0.002), consistent with the finding from Paper 1 that these transition economies are influential outliers.

### 9.2 Regional Subsamples

The OECD subsample shows no significant demographic effect on Japanification (Z₁ = −0.35, p = 0.73; all Z p > 0.73). The Asia subsample shows a *negative* significant Z₁ (−1.72, p = 0.042), meaning aging in Asia is associated with less Japanification --- likely reflecting rapid growth catch-up that offsets demographic headwinds. The Europe subsample shows no significant demographic effect (Z₁ = 0.50, p = 0.71).

### 9.3 Alternative Indices

The rolling 5-year moving average index produces the strongest demographic result (Z₁ = 1.97, p < 0.001; all Z p < 0.001), consistent with the interpretation that Japanification is a persistent state and year-to-year fluctuations introduce noise that weakens the signal. The PCA-based index shows no significant demographic effect (Z₁ = −0.15, p = 0.86), as the first principal component loads primarily on inflation rather than growth. The variance-weighted index produces identical results to the equal-weighted version due to similar component variances.

### 9.4 Goodhart-Pradhan Reversal

We find no evidence that the demographic-inflation relationship has reversed sign post-2020. On raw annual inflation (1990--2019), demographics are marginally significant in the *disinflationary* direction (Z₁ = −50.1, p = 0.058), the opposite of what Goodhart-Pradhan predict. In advanced economies, demographics are completely null on inflation (Z₁ = −10.8, p = 0.63); in emerging markets, the disinflationary signal is marginally significant (Z₁ = −67.3, p = 0.063). The Juselius-Takáts U-shape regression produces correctly signed coefficients for youth dependency (+), old-age dependency (+), and working-age share (+), but none are statistically significant. If aging is becoming inflationary as Goodhart and Pradhan predict, the effect has not yet materialized in the cross-country data.


## 10. Discussion and Conclusion

### 10.1 Demographics Do Not Predict Japanification

The central finding of this paper is a qualified one: demographics weakly predict "Japanification" on the full sample (Z₁ = 1.179, p = 0.003), but the result is deeply unstable. It reverses sign post-GFC, disappears in high-income countries, and is substantially attenuated by working-age share controls in low-income countries. The canonical Japanification syndrome requires three simultaneous conditions --- low growth, low inflation, and low rates --- and the demographic signal, to the extent it exists, does not robustly predict any of them across all specifications, periods, and samples.

The component decomposition resolves the channel question. In the chronic/rolling definition that matches the Japanification concept, the persistent demographic signal --- where present --- operates primarily through disinflation, not through growth. All three Z terms predict lower inflation on five-year rolling averages (Z₁ = −3.25, p < 0.001), while no Z term significantly predicts growth (Z₁ p = 0.56). In raw annual data, demographics sometimes correlate with growth, but that relationship is transitory and washes out under rolling-average smoothing. On a homogeneous 10-year yield subsample, demographics do predict interest rates --- but only pre-GFC. No single channel is robustly significant across all specifications, periods, and samples.

This reframing matters because it redirects the policy conversation. If Japanification were a demographic inevitability, the appropriate response would be resignation or extreme policy intervention. But the evidence suggests demographics produce a persistent disinflationary tendency (Z₁ = −3.25 on rolling inflation) and a transitory growth association concentrated in middle-income economies losing their demographic dividend --- not the full monetary-price syndrome that defines Japan's post-bubble experience. The strengthened baseline (p = 0.003 versus the prior marginal p = 0.049) confirms that demographics matter, but the channel through which they matter --- disinflation, not growth --- and the post-GFC sign reversal mean the policy challenge is more conventional: maintaining growth through productivity-enhancing reforms, automation investment, and immigration policy.

Japan's actual experience likely involved demographics amplifying country-specific factors (the asset bubble aftermath, the banking crisis, institutional rigidities, cultural attitudes toward risk and entrepreneurship) rather than demographics alone generating the full syndrome. The cross-country evidence does not support aging as a sufficient condition for Japanification.

### 10.2 The Inflation Puzzle and the Goodhart-Pradhan Thesis

The null result for demographics on inflation deserves explicit discussion because it contradicts two competing theoretical predictions. The traditional lifecycle framework (Juselius and Takáts, 2015) predicts that high dependency ratios (both young and old) are inflationary because dependents consume without producing. Goodhart and Pradhan (2020) argue more strongly that aging is inflationary through labor scarcity: as workers become scarce, wages rise, pushing up costs and prices. Both predict a positive relationship between aging and inflation.

Our results are more nuanced than a simple null. On raw year-to-year data, demographics are marginally significant in the disinflationary direction (Z₁ = −50.1 on inflation, p = 0.058). On five-year rolling averages --- which capture the persistent component relevant to the Japanification concept --- all three Z terms are highly significant predictors of inflation (Z₁ = −3.25, p < 0.001). The sign is disinflationary: aging demographics predict *lower* inflation, consistent with the standard lifecycle savings channel and inconsistent with the Goodhart-Pradhan reversal. Demographics do have a persistent effect on inflation, but it runs in exactly the opposite direction from what Goodhart and Pradhan predict. If aging is becoming inflationary through labor scarcity, the effect has not materialized in three decades of cross-country data covering the most dramatic aging episodes in history.

One resolution is that the inflationary effect of aging operates with a much longer lag than the growth effect. Aging depresses growth immediately (through labor supply) but only becomes inflationary once labor scarcity actually bites --- which requires tight labor markets, institutional wage-setting power, and the absence of automation and immigration as substitutes. Our companion paper on automation (Paper J) shows that aging economies adopt capital-intensive technologies, which may offset the labor scarcity channel and prevent the Goodhart-Pradhan reversal from materializing. Another resolution is that globalization and anchored inflation expectations have suppressed the inflation channel during our sample period, and that the Goodhart-Pradhan thesis may eventually prove correct as these countervailing forces weaken.

### 10.3 The Structural Break and Policy

The structural break is more dramatic than a mere "disappearance" --- the Z₁ coefficient reverses sign from +2.89 (p < 0.001) pre-GFC to −1.46 (p < 0.001) post-GFC on the clean rolling index. This sign reversal survives balanced-panel, OECD/non-OECD, income-quartile, and global-factor controls, ruling out entry/exit of countries as the primary explanation. However, the QE split (Section 6.4) shows that QE and non-QE countries had opposite Z₁ signs even before the crisis, suggesting the break is linked to policy-regime heterogeneity that was already present pre-GFC rather than a discrete structural change alone. The break is not confined to rich economies: it appears in both OECD and non-OECD subsamples, though the pre-GFC significance was driven primarily by non-OECD countries.

Multiple interpretations remain viable: global contagion making Japanification a policy-transmitted rather than demographically driven condition; a convergence-era artifact that dissipated as transition economies matured; or a policy regime change in which unconventional monetary policy overrode demographic signals. Our data cannot definitively adjudicate among these. The pre-GFC relationship was real and robust. Whether it was a permanent feature of economic structure or a feature of a specific historical era remains an open question.

### 10.4 The Demographic Dividend Connection

The finding that the Japanification effect operates through working-age share rather than old-age dependency has important implications. It connects Japanification not to aging per se but to the broader demographic transition. Countries experience their demographic dividend when the working-age share is rising; when it peaks and declines, growth slows and the Japanification index rises mechanically. This is a universal feature of the demographic transition, but it has different implications than the aging narrative.

Specifically, it suggests that countries like India and Nigeria --- currently benefiting from rising working-age shares --- will eventually face Japanification-like growth slowdowns as their transitions mature, regardless of whether they develop Japan-style institutional features. Conversely, for countries that have already completed the transition (Japan, Germany, Italy), the working-age share effect has already materialized, and further aging may not produce additional Japanification --- consistent with our finding that the demographic effect is absent in high-income countries.

### 10.5 Implications for Projections

Our conditional projections expose the fundamental fragility of demographic Japanification forecasts. Under pre-GFC coefficients, 10 of 24 focus economies cross the threshold by 2050. Under post-GFC or full-sample coefficients, none do. The divergence is not marginal --- Japan's projected 2050 index is 1.08 under pre-GFC coefficients but 0.11 under post-GFC coefficients, a factor-of-ten difference.

This extreme sensitivity to estimation period means that any single-scenario projection would be misleading. The appropriate interpretation is that demographic projections of Japanification are unreliable: they depend entirely on whether the pre-GFC relationship --- which may have captured convergence dynamics rather than a permanent demographic-macro link --- continues to hold.

### 10.6 Conclusion

Demographics weakly predict Japanification on the full sample (Z₁ = 1.179, p = 0.003), but this association masks deep instabilities: the coefficient reverses sign post-GFC, is concentrated in middle-income countries losing their demographic dividend, and is substantially attenuated by working-age share controls. In the chronic/rolling definition that matches the Japanification concept, the persistent demographic signal operates through disinflation (Z₁ = −3.25, p < 0.001 on rolling inflation), not through growth (Z₁ p = 0.56 on rolling growth). The rate channel is significant only on homogeneous 10-year yield data and only pre-GFC. Conditional projections under different estimation periods diverge by a factor of ten, rendering single-scenario forecasts unreliable. The most honest summary is that demographics contribute a persistent disinflationary tendency and contributed to pre-GFC Japanification dynamics in converging economies, but whether the broader relationship survives in the post-GFC world --- where policy, institutions, and global contagion dominate --- is the central open question for researchers and policymakers confronting an aging world.


## Appendix A: Homogeneous Rate Measure

The rate hierarchy employed in the main panel (Section 3.1) mixes 10-year government bond yields, policy rates, and lending rates, raising the concern that rate channel results are contaminated by measurement heterogeneity. This appendix restricts to the 23 countries with 10-year government bond yield data (635 observations) and constructs a homogeneous three-component Japanification index.

**Baseline.** On the homogeneous 3c index (635 observations, 23 countries), all three demographic polynomials are highly significant (Z₁ = −8.94, p < 0.001; Z₂ = 1.08, p = 0.001; Z₃ = −0.036, p = 0.004), with R² = 0.294. The negative Z₁ sign means that aging demographics predict *less* Japanification on the 3c index in this restricted sample --- reflecting that the rate channel moves in a direction that partially offsets the growth/inflation channels.

**Pre/post GFC.** The break is present here as well. Pre-GFC (359 observations): Z₁ = −10.87 (p < 0.001), R² = 0.326. Post-GFC (253 observations): all Z terms are null (Z₁ = −0.64, p = 0.79; Z₂ p = 0.83; Z₃ p = 0.87), R² = 0.442.

**Direct rate regression.** Z → 10-year government bond yield: Z₁ = 53.1 (p = 0.011), Z₂ = −5.96 (p = 0.031), Z₃ = 0.19 (p = 0.071). Demographics *do* predict interest rates when measured homogeneously, with the expected sign (aging → lower yields). On the heterogeneous rate measure (rate_japan, 3,041 observations), Z terms are actually significant but with the opposite sign (Z₁ = −43.9, p = 0.010) --- aging predicts *lower* heterogeneous rates, which may reflect the splicing of bond yields, policy rates, and lending rates across different country types.

**Conclusion.** On homogeneous 10-year yield data, demographics significantly predict interest rates --- but only pre-GFC. The rate channel exists but is historically contingent, paralleling the structural break in the composite index.


## References

Aksoy, Y., Basso, H. S., Smith, R. P., and Grasl, T. (2019). Demographic structure and macroeconomic trends. *American Economic Journal: Macroeconomics*, 11(1), 193--222.

Bernanke, B. S. (2005). The global saving glut and the U.S. current account deficit. *Sandridge Lecture*, Virginia Association of Economists.

Bouis, R., Cournède, B., and Christensen, A. K. (2013). Implications of output gap uncertainty in times of crisis. *OECD Economics Department Working Papers*, No. 1053.

Carvalho, C., Ferrero, A., and Nechio, F. (2016). Demographics and real interest rates: Inspecting the mechanism. *European Economic Review*, 88, 208--226.

Fair, R. C. and Dominguez, K. M. (1991). Effects of the changing U.S. age distribution on macroeconomic equations. *American Economic Review*, 81(5), 1276--1294.

Goodhart, C. A. E. and Pradhan, M. (2020). *The Great Demographic Reversal: Ageing Societies, Waning Inequality, and an Inflation Revival*. Palgrave Macmillan.

Higgins, M. (1998). Demography, national savings, and international capital flows. *International Economic Review*, 39(2), 343--369.

Juselius, M. and Takáts, E. (2015). Can demography affect inflation and monetary policy? *BIS Working Papers*, No. 485.

Koomen, M. and Wicht, L. (2020). Demographics and current account imbalances. *IMF Working Paper*, WP/20/225.

Modigliani, F. and Brumberg, R. (1954). Utility analysis and the consumption function: An interpretation of cross-section data. In K. K. Kurihara (Ed.), *Post-Keynesian Economics*, pp. 388--436. Rutgers University Press.

Rachel, L. and Summers, L. H. (2019). On secular stagnation in the industrialized world. *Brookings Papers on Economic Activity*, Spring.

Summers, L. H. (2014). U.S. economic prospects: Secular stagnation, hysteresis, and the zero lower bound. *Business Economics*, 49(2), 65--73.
